Перед вами датасет с данными о вреде курения

Структура данных:

1. gender: Пол пациента
2. age: Возраст пациента
3. hypertension: Наличие гипертонии у пациента 1 - есть, 0 - не наблюдалось
4. heart_disease: Наличие у пациента болезней связанных с сердцем 1 - есть, 0 - здоров
5. ever_married: Брак - (Состоял в браке/ холост)
6. work_type: Работ - Ребенок(children), Госслужащий(Govt_jov), Никогда не работал(Never_worked), Самозанятый(self-employed) или Скрыто(private)
7. Residence_type: место работы городская среда(Urban)/удаленка или за городом(Rural)
8. avg_glucose_level: средний уровень глюкозы в крови
9. bmi: индекс массы тела
10. smoking_status: Пациент курит(Иногда, всегда, время от времени или Unknown - неизвестно)
11. stroke: Перенесенный инсульт (1 - пациент перенес инсульт, 0 - инсульта не было) - ваш target

Домашнее задание №2

Ваша задача

№1 Восстановить целостностью данных, реализовать функцию позволяющие выполнять заполнение данных по среднему или по медианному значению в Series.

1. Заполнить пропуски
2. Убрать дубликаты
3. Убрать некорректные данные и.т.п.

№2 Построить baseline-модель, попытаться выбить лучший скор (сравнить несколько моделей)

№3 Сделать визуализацию для мини-исследования о вреде курения с разбивкой по половой принадлежности пациентов с помощью pandas, matplotlib, seaborn, plotly и других инструментов.

import pandas as pd
import numpy as np
import seaborn as sns
import math

url = 'https://drive.google.com/file/d/1He5GI5_Gd8uXYfeETLBISQ5BszX0o4pU/view?usp=sharing'
url = 'https://drive.google.com/uc?id=' + url.split('/')[-2]
data = pd.read_csv(url)
data = data.drop('Unnamed: 0',axis=1)
data = data.drop(0,axis=0)
data = data.sample(frac=1)
data = data.reset_index(drop=True)

data.head(10)

	gender	age	hypertension	heart_disease	ever_married	work_type	Residence_type	avg_glucose_level	bmi	smoking_status	stroke
0	Female	43.0	0.0	0.0	Yes	Private	Rural	58.63	28.4	smokes	0
1	Female	29.0	0.0	0.0	Yes	Private	Rural	72.52	33.9	never smoked	0
2	Female	40.0	0.0	NaN	Yes	Govt_job	Rural	110.6	33.3	formerly smoked	0
3	Female	32.0	0.0	0.0	No	Self-employed	Rural	93.17	27.5	smokes	0
4	Female	3.0	0.0	0.0	No	children	Rural	97.6	25.8	Unknown	0
5	Female	20.0	0.0	0.0	No	Private	Rural	93.74	23.7	Unknown	0
6	Female	34.0	0.0	0.0	Yes	Govt_job	Rural	70.18	24.9	Unknown	0
7	Female	51.0	1.0	0.0	Yes	Private	Urban	109.16	28.0	smokes	0
8	Female	79.0	0.0	0.0	Yes	NaN	Urban	57.77	NaN	formerly smoked	0
9	Female	46.0	1.0	0.0	Yes	Private	Rural	81.58	36.2	never smoked	0

def my_fill_na_avg(df, columnName):
    
    columnsType = {'gender':'str', 'age':'int', 'hypertension':'int', 'heart_disease':'int', 'ever_married':'str', 
                   'work_type':'str', 'Residence_type':'str', 'avg_glucose_level':'float', 'bmi':'float', 
                   'smoking_status':'str', 'stroke':'int'}
    if not columnName in columnsType:
        return -1
    else:
        col = df[columnName]
        sum = 0
        countObj = 0
        if (columnsType[columnName] == 'int'):
            for row in col:
                if not pd.isna(row):
                    sum += int(row.split('.')[0])
                    countObj += 1
            avg = sum / countObj if countObj != 0 else -1
            if avg == -1:
                return col
            count = 0
            for row in col:
                if pd.isna(row):
                    col[count] = int(round(avg))
                else:
                    col[count] = int(row.split('.')[0])
                count += 1

        if (columnsType[columnName] == 'float'):
            for row in col:
                if not pd.isna(row):
                    sum += float(row)
                    countObj += 1
            avg = sum / countObj if countObj != 0 else -1
            if avg == -1:
                return col
            count = 0
            for row in col:
                if pd.isna(row):
                    col[count] = round(float(avg),2)
                else:
                    col[count] = round(float(row),2)
                count += 1

        if (columnsType[columnName] == 'str'):
            avgStr = dict()
            for row in col:
                #выберем часто встречающийся объект, т.к. невозможно найти среднее между string
                if not pd.isna(row):
                    count = avgStr[row] if row in avgStr.keys() else 1
                    avgStr[row] = count+1
                    
            max_val = max(avgStr.values())
            avg = [k for k, v in avgStr.items() if v == max_val]

            count = 0
            for row in col:
                if pd.isna(row):
                    col[count] = avg[0]
                else:
                    col[count] = row
                count += 1
    return col

data_avg = data.copy()
for col in data.columns:
    data_avg[col] = my_fill_na_avg(data_avg, col)
data_avg.head(10)

	gender	age	hypertension	heart_disease	ever_married	work_type	Residence_type	avg_glucose_level	bmi	smoking_status	stroke
0	Female	43	0	0	Yes	Private	Rural	58.63	28.4	smokes	0
1	Female	29	0	0	Yes	Private	Rural	72.52	33.9	never smoked	0
2	Female	40	0	0	Yes	Govt_job	Rural	110.6	33.3	formerly smoked	0
3	Female	32	0	0	No	Self-employed	Rural	93.17	27.5	smokes	0
4	Female	3	0	0	No	children	Rural	97.6	25.8	Unknown	0
5	Female	20	0	0	No	Private	Rural	93.74	23.7	Unknown	0
6	Female	34	0	0	Yes	Govt_job	Rural	70.18	24.9	Unknown	0
7	Female	51	1	0	Yes	Private	Urban	109.16	28.0	smokes	0
8	Female	79	0	0	Yes	Private	Urban	57.77	28.94	formerly smoked	0
9	Female	46	1	0	Yes	Private	Rural	81.58	36.2	never smoked	0

def my_fill_na_mediana(df, columnName):
    
    columnsType = {'gender':'str', 'age':'int', 'hypertension':'int', 'heart_disease':'int', 'ever_married':'str', 
                   'work_type':'str', 'Residence_type':'str', 'avg_glucose_level':'float', 'bmi':'float', 
                   'smoking_status':'str', 'stroke':'int'}
    if not columnName in columnsType:
        return -1
    else:
        col = df[columnName]
        median = 0
        medList = list()
        if (columnsType[columnName] == 'int'):
            for row in col:
                if not pd.isna(row):
                    medList.append(int(row.split('.')[0]))
            medList.sort()
            
            if len(medList) % 2 == 0:
                median = int((medList[int(len(medList)/2-1)]+medList[int(len(medList)/2)])/2)
            else:
                median = int(medList[math.floor(len(medList)/2)])
            count = 0
            for row in col:
                if pd.isna(row):
                    col[count] = int(round(median))
                else:
                    col[count] = int(row.split('.')[0])
                count += 1
                
        if (columnsType[columnName] == 'float'):
            for row in col:
                if not pd.isna(row):
                    medList.append(float(row))
            medList.sort()
            
            if len(medList) % 2 == 0:
                median = float((medList[int(len(medList)/2-1)]+medList[int(len(medList)/2)])/2)
            else:
                median = float(medList[math.floor(len(medList)/2)])
            count = 0
            for row in col:
                if pd.isna(row):
                    col[count] = float(round(median))
                else:
                    col[count] = float(row)
                count += 1
                
        if (columnsType[columnName] == 'str'):
            for row in col:
                if not pd.isna(row):
                    medList.append(str(row))
            medList.sort()

            #т.к. невозможно выбрать среднее между двумя string
            median = str(medList[math.floor(len(medList)/2)])
            count = 0
            for row in col:
                if pd.isna(row):
                    col[count] = str(median)
                else:
                    col[count] = str(row)
                count += 1
    return col

data_mediana = data.copy()
for col in data.columns:
    data_mediana[col] = my_fill_na_mediana(data_mediana, col)
data_mediana.head(10)

	gender	age	hypertension	heart_disease	ever_married	work_type	Residence_type	avg_glucose_level	bmi	smoking_status	stroke
0	Female	43	0	0	Yes	Private	Rural	58.63	28.4	smokes	0
1	Female	29	0	0	Yes	Private	Rural	72.52	33.9	never smoked	0
2	Female	40	0	0	Yes	Govt_job	Rural	110.6	33.3	formerly smoked	0
3	Female	32	0	0	No	Self-employed	Rural	93.17	27.5	smokes	0
4	Female	3	0	0	No	children	Rural	97.6	25.8	Unknown	0
5	Female	20	0	0	No	Private	Rural	93.74	23.7	Unknown	0
6	Female	34	0	0	Yes	Govt_job	Rural	70.18	24.9	Unknown	0
7	Female	51	1	0	Yes	Private	Urban	109.16	28.0	smokes	0
8	Female	79	0	0	Yes	Private	Urban	57.77	28.0	formerly smoked	0
9	Female	46	1	0	Yes	Private	Rural	81.58	36.2	never smoked	0

Удалим полные дубликаты

data_avg.drop_duplicates(ignore_index=True)
data_mediana.drop_duplicates(ignore_index=True)

	gender	age	hypertension	heart_disease	ever_married	work_type	Residence_type	avg_glucose_level	bmi	smoking_status	stroke
0	Female	43	0	0	Yes	Private	Rural	58.63	28.4	smokes	0
1	Female	29	0	0	Yes	Private	Rural	72.52	33.9	never smoked	0
2	Female	40	0	0	Yes	Govt_job	Rural	110.6	33.3	formerly smoked	0
3	Female	32	0	0	No	Self-employed	Rural	93.17	27.5	smokes	0
4	Female	3	0	0	No	children	Rural	97.6	25.8	Unknown	0
...	...	...	...	...	...	...	...	...	...	...	...
5101	Male	1	0	0	No	children	Rural	78.53	19.8	Unknown	0
5102	Male	65	0	0	Yes	Private	Rural	236.14	28.0	Unknown	0
5103	Female	50	0	0	Yes	Self-employed	Urban	110.18	26.0	formerly smoked	0
5104	Male	81	0	1	Yes	Self-employed	Rural	68.27	25.0	Unknown	0
5105	Male	65	1	0	Yes	Self-employed	Urban	113.86	36.4	never smoked	0

5106 rows × 11 columns

Преобразуем столбцы в правильный тип данных

columnsType = {'gender':'str', 'age':'int', 'hypertension':'int', 'heart_disease':'int', 'ever_married':'str', 
                   'work_type':'str', 'Residence_type':'str', 'avg_glucose_level':'float', 'bmi':'float', 
                   'smoking_status':'str', 'stroke':'int'}

for col in data.columns:
    type = columnsType[col]
    data_avg[col] = data_avg[col].astype(str).astype(type)
    data_mediana[col] = data_mediana[col].astype(str).astype(type)

print(data_avg.info())
print(data_mediana.info())

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 5110 entries, 0 to 5109
Data columns (total 11 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   gender             5110 non-null   object 
 1   age                5110 non-null   int32  
 2   hypertension       5110 non-null   int32  
 3   heart_disease      5110 non-null   int32  
 4   ever_married       5110 non-null   object 
 5   work_type          5110 non-null   object 
 6   Residence_type     5110 non-null   object 
 7   avg_glucose_level  5110 non-null   float64
 8   bmi                5110 non-null   float64
 9   smoking_status     5110 non-null   object 
 10  stroke             5110 non-null   int32  
dtypes: float64(2), int32(4), object(5)
memory usage: 359.4+ KB
None
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 5110 entries, 0 to 5109
Data columns (total 11 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   gender             5110 non-null   object 
 1   age                5110 non-null   int32  
 2   hypertension       5110 non-null   int32  
 3   heart_disease      5110 non-null   int32  
 4   ever_married       5110 non-null   object 
 5   work_type          5110 non-null   object 
 6   Residence_type     5110 non-null   object 
 7   avg_glucose_level  5110 non-null   float64
 8   bmi                5110 non-null   float64
 9   smoking_status     5110 non-null   object 
 10  stroke             5110 non-null   int32  
dtypes: float64(2), int32(4), object(5)
memory usage: 359.4+ KB
None

import matplotlib.pyplot as plt
plt.figure(figsize=(20, 10))
plt.plot(data_avg['gender'])
plt.show()

#Удалим выброс
data_avg = data_avg.loc[data_avg['gender'] != 'Other']
data_avg = data_avg.reset_index(drop=True)

from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import AdaBoostClassifier
from sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OrdinalEncoder

Обучим модель рандомного леса на данных, у которых пропуски заменены на средние значения

Преобразуем строковый тип в числовой с помощью OrdinalEncoder

features = data_avg.drop("stroke", axis=1)
target = data_avg["stroke"]

bad_series = features[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']]
features_oe = features.copy()
encoder = OrdinalEncoder()
features_oe[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']] = pd.DataFrame(encoder.fit_transform(bad_series), columns=bad_series.columns)
features_oe.head()

	gender	age	hypertension	heart_disease	ever_married	work_type	Residence_type	avg_glucose_level	bmi	smoking_status
0	0.0	43	0	0	1.0	2.0	0.0	58.63	28.4	3.0
1	0.0	29	0	0	1.0	2.0	0.0	72.52	33.9	2.0
2	0.0	40	0	0	1.0	0.0	0.0	110.60	33.3	1.0
3	0.0	32	0	0	0.0	3.0	0.0	93.17	27.5	3.0
4	0.0	3	0	0	0.0	4.0	0.0	97.60	25.8	0.0

train, test, target_train, target_test = train_test_split(
    features_oe, target, test_size=0.25)

rfc = RandomForestClassifier(random_state=123)
model_params = {
            "n_estimators"      : [10,20,30],
            "max_features"      : ["auto", "sqrt", "log2"],
            "min_samples_split" : [2,4,8],
            "bootstrap": [True, False],}
grid_rfc = GridSearchCV(rfc, model_params)

grid_rfc.fit(train, target_train)
print(grid_rfc.best_params_)
pred_log = grid_rfc.predict(test)

f_1_log = f1_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
precision = precision_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
recall = recall_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
accuracy_log = accuracy_score(target_test, pred_log)
print('Значение F1 = {}, \nзначение precision = {}, \nзначение recall = {}, \nзначение accuracy = {}'.format(f_1_log, precision, recall, accuracy_log))

{'bootstrap': True, 'max_features': 'auto', 'min_samples_split': 4, 'n_estimators': 30}
Значение F1 = 0.950876496734705, 
значение precision = 0.9457362913866642, 
значение recall = 0.9655712050078247, 
значение accuracy = 0.9655712050078247

Обучим модель рандомного леса на данных, у которых пропуски заменены на медианой

features = data_mediana.drop("stroke", axis=1)
target = data_mediana["stroke"]

bad_series = features[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']]
features_oe = features.copy()
encoder = OrdinalEncoder()
features_oe[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']] = pd.DataFrame(encoder.fit_transform(bad_series), columns=bad_series.columns)

train, test, target_train, target_test = train_test_split(
    features_oe, target, test_size=0.25)

rfc = RandomForestClassifier(random_state=123)
model_params = {
            "n_estimators"      : [10,20,30],
            "max_features"      : ["auto", "sqrt", "log2"],
            "min_samples_split" : [2,4,8],
            "bootstrap": [True, False],}
grid_rfc = GridSearchCV(rfc, model_params)

grid_rfc.fit(train, target_train)
print(grid_rfc.best_params_)
pred_log = grid_rfc.predict(test)

f_1_log = f1_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
precision = precision_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
recall = recall_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
accuracy_log = accuracy_score(target_test, pred_log)
print('Значение F1 = {}, \nзначение precision = {}, \nзначение recall = {}, \nзначение accuracy = {}'.format(f_1_log, precision, recall, accuracy_log))

{'bootstrap': True, 'max_features': 'auto', 'min_samples_split': 8, 'n_estimators': 20}
Значение F1 = 0.9320605697102562, 
значение precision = 0.9112589046853995, 
значение recall = 0.9538341158059468, 
значение accuracy = 0.9538341158059468

Обучим модель адаптивного бустинга на данных, у которых пропуски заменены на средние значения

features = data_avg.drop("stroke", axis=1)
target = data_avg["stroke"]

bad_series = features[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']]
features_oe = features.copy()
encoder = OrdinalEncoder()
features_oe[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']] = pd.DataFrame(encoder.fit_transform(bad_series), columns=bad_series.columns)

train, test, target_train, target_test = train_test_split(
    features_oe, target, test_size=0.25)

abc = AdaBoostClassifier(random_state=123)
model_params = {
                "n_estimators" : [10, 20, 30, 40],
                "learning_rate" : [0.2, 0.5, 0.7, 1]
             }
grid_abc = GridSearchCV(abc, model_params)

grid_abc.fit(train, target_train)
print(grid_abc.best_params_)
pred_log = grid_abc.predict(test)

f_1_log = f1_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
precision = precision_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
recall = recall_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
accuracy_log = accuracy_score(target_test, pred_log)
print('Значение F1 = {}, \nзначение precision = {}, \nзначение recall = {}, \nзначение accuracy = {}'.format(f_1_log, precision, recall, accuracy_log))

{'learning_rate': 0.7, 'n_estimators': 20}
Значение F1 = 0.9788254095085897, 
значение precision = 0.958528951486698, 
значение recall = 1.0, 
значение accuracy = 0.9585289514866979

Обучим модель адаптивного бустинга на данных, у которых пропуски заменены на медианой

features = data_mediana.drop("stroke", axis=1)
target = data_mediana["stroke"]

bad_series = features[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']]
features_oe = features.copy()
encoder = OrdinalEncoder()
features_oe[['gender', 'ever_married', 'work_type', 'Residence_type', 'smoking_status']] = pd.DataFrame(encoder.fit_transform(bad_series), columns=bad_series.columns)

train, test, target_train, target_test = train_test_split(
    features_oe, target, test_size=0.25)

abc = AdaBoostClassifier(random_state=123)
model_params = {
                "n_estimators" : [10, 20, 30, 40],
                "learning_rate" : [0.2, 0.5, 0.7, 1]
             }
grid_abc = GridSearchCV(abc, model_params)

grid_abc.fit(train, target_train)
print(grid_abc.best_params_)
pred_log = grid_abc.predict(test)

f_1_log = f1_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
precision = precision_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
recall = recall_score(target_test, pred_log, average='weighted', labels=np.unique(pred_log))
accuracy_log = accuracy_score(target_test, pred_log)
print('Значение F1 = {}, \nзначение precision = {}, \nзначение recall = {}, \nзначение accuracy = {}'.format(f_1_log, precision, recall, accuracy_log))

{'learning_rate': 0.2, 'n_estimators': 10}
Значение F1 = 0.9743178170144463, 
значение precision = 0.9499217527386542, 
значение recall = 1.0, 
значение accuracy = 0.9499217527386542

В итоге обе модели показали лучший рещультат на данных, в которых пропуски заполнены с помощью среднего значения. В целом точнее оказалась модель адаптивного бустинга.

Мини исследование

#Удаление срок с пустыми данными сократит датасет в два раза, поэтому возьмем все данные. 
#Исправленые датасеты тоже не будет брать, потому что они значительно изменят достоверность статистики.
df = data.copy()

#Разделим датафрейм на два. В первом находятся данные только о мужчинах, во втором только о женщинах.
df_male = df.loc[df['gender'] == 'Male']
df_male = df_male.reset_index(drop=True)

df_female = df.loc[df['gender'] == 'Female']
df_female = df_female.reset_index(drop=True)

Smoking status у людей с гипертонией

male_smoking_hypertension = dict()
for obj in df_male.iterrows():
        if obj[1]['hypertension'] == '1.0':
            row = obj[1]['smoking_status']
            count = male_smoking_hypertension[row] if row in male_smoking_hypertension.keys() else 1
            male_smoking_hypertension[row] = count+1   
print(male_smoking_hypertension)

female_smoking_hypertension = dict()
for obj in df_female.iterrows():
        if obj[1]['hypertension'] == '1.0':
            row = obj[1]['smoking_status']
            count = female_smoking_hypertension[row] if row in female_smoking_hypertension.keys() else 1
            female_smoking_hypertension[row] = count+1   
print(female_smoking_hypertension)

{'smokes': 47, 'formerly smoked': 59, 'never smoked': 85, 'Unknown': 25}
{'smokes': 46, 'never smoked': 136, 'Unknown': 26, 'formerly smoked': 57}

plt.title('Smoking status у мужчин с гипертонией')
plt.xlabel('Smoking status')
plt.ylabel('Количество мужчин', rotation=90)
names = list(male_smoking_hypertension.keys())
values = list(male_smoking_hypertension.values())
plt.bar(range(len(male_smoking_hypertension)), values, tick_label=names)
plt.show()

plt.title('Smoking status у женщин с гипертонией')
plt.xlabel('Smoking status')
plt.ylabel('Количество женщин', rotation=90)
names = list(female_smoking_hypertension.keys())
values = list(female_smoking_hypertension.values())
plt.bar(range(len(female_smoking_hypertension)), values, tick_label=names)
plt.show()

Видим, что больше всего мужчин с гипертонией которые никогда не курили, а меньше которые курят на регулярной основе. То же самое можно наблюдать у женщин.

Smoking status у людей с болезнью сердца

male_smoking_heart_disease = dict()
for obj in df_male.iterrows():
        if obj[1]['heart_disease'] == '1.0':
            row = obj[1]['smoking_status']
            count = male_smoking_heart_disease[row] if row in male_smoking_heart_disease.keys() else 1
            male_smoking_heart_disease[row] = count+1   
print(male_smoking_heart_disease)

female_smoking_heart_diseasen = dict()
for obj in df_female.iterrows():
        if obj[1]['heart_disease'] == '1.0':
            row = obj[1]['smoking_status']
            count = female_smoking_heart_diseasen[row] if row in female_smoking_heart_diseasen.keys() else 1
            female_smoking_heart_diseasen[row] = count+1   
print(female_smoking_heart_diseasen)

{'never smoked': 36, 'formerly smoked': 45, 'Unknown': 25, 'smokes': 31}
{'never smoked': 38, 'smokes': 22, 'formerly smoked': 24, 'Unknown': 17}

import plotly.express as px
names = list(male_smoking_heart_disease.keys())
values = list(male_smoking_heart_disease.values())
fig = px.pie(values=values, names=names, title='Smoking status у мужчин с болезнью сердца')
fig.update_traces(textinfo='percent+label')
fig.show()

names = list(female_smoking_heart_diseasen.keys())
values = list(female_smoking_heart_diseasen.values())
fig = px.pie(values=values, names=names, title='Smoking status у женщин с болезнью сердца')
fig.update_traces(textinfo='percent+label')
fig.show()

Видим, что болезни, связанные с сердцем, чаще бывают у мужчин, которые раньше курили, а реже которые курят на регулярной основе. У женщин чаще бывают у тех, которые никогда не курили, реже которые курят на регулярной основе.

Smoking status у людей с перенесенным инсультом

male_smoking_stroke = dict()
for obj in df_male.iterrows():
        if obj[1]['stroke'] == '1':
            row = obj[1]['smoking_status']
            count = male_smoking_stroke[row] if row in male_smoking_stroke.keys() else 1
            male_smoking_stroke[row] = count+1   
print(male_smoking_stroke)

female_smoking_stroke = dict()
for obj in df_female.iterrows():
        if obj[1]['stroke'] == '1':
            row = obj[1]['smoking_status']
            count = female_smoking_stroke[row] if row in female_smoking_stroke.keys() else 1
            female_smoking_stroke[row] = count+1   
print(female_smoking_stroke)

{'never smoked': 28, 'formerly smoked': 37, 'Unknown': 23, 'smokes': 24}
{'never smoked': 64, 'Unknown': 26, 'formerly smoked': 35, 'smokes': 20}

names1 = list(male_smoking_stroke.keys())
values1 = list(male_smoking_stroke.values())

names2 = names1
values2 = [female_smoking_stroke[names2[0]], female_smoking_stroke[names2[1]],
              female_smoking_stroke[names2[2]], female_smoking_stroke[names2[3]]]

plt.figure(figsize=(10, 5))
plt.title('Smoking status у людей с перенесенным инсультом')
plt.xlabel('Smoking status')
plt.ylabel('Количество', rotation=90)
plt.plot(names1, values1, "-b", label="Male")
plt.plot(names2, values2, "-r", label="Female")
plt.legend(loc="upper left")
plt.show()

Видим, что инсульт, чаще бывают у мужчин, которые раньше курили, а реже которые курят на регулярной основе. У женщин чаще бывают у тех, которые никогда не курили, реже которые курят на регулярной основе.

Распределение курящих людей по возрасту

age_smoking = dict()
for obj in df_male.iterrows():
        if obj[1]['smoking_status'] == 'smokes' and not pd.isna(obj[1]['age']):
            row = str(obj[1]['gender']) + " " + str(obj[1]['age'])
            count = age_smoking[row] if row in age_smoking.keys() else 1
            age_smoking[row] = count+1   

for obj in df_female.iterrows():
        if obj[1]['smoking_status'] == 'smokes' and not pd.isna(obj[1]['age']):
            row = str(obj[1]['gender']) + " " + str(obj[1]['age'])
            count = age_smoking[row] if row in age_smoking.keys() else 1
            age_smoking[row] = count+1   

df_age_smoking = pd.DataFrame(columns = ['gender', 'age', 'count'])
for item in age_smoking.items():
    gender, age = item[0].split(" ")
    gender, age = str(gender), float(age)
    row = pd.DataFrame([{'gender' : gender, 'age' : age, 'count' : item[1]}])
    df_age_smoking = pd.concat([df_age_smoking, row], ignore_index=True)

import seaborn as sns
sns.set_theme(style="whitegrid")

cmap = sns.cubehelix_palette(rot=-.2, as_cmap=True)
g = sns.relplot(
    data=df_age_smoking,
    x="age", y="gender",  size="count",
    palette=cmap, sizes=(20, 200), height=5, aspect=3,
)
g.ax.xaxis.grid(True, "minor", linewidth=.25)
g.ax.yaxis.grid(True, "minor", linewidth=.25)
g.despine(left=True, bottom=True)

<seaborn.axisgrid.FacetGrid at 0x1c743bf0eb0>

Видим, что по статистике больше всего курящих мужчин в возрасте от 25 до 32 и от 48 до 57 лет.
У женщин не видно таких явных пиков, но если посмотреть внимательней, то можно сказать что больше курящих женщин в районе 30 и от 40 до 55 лет. По графику видно, что курящих девушек больше, чем курящих мужчин.

В итоге можно сделать вывод о том, что заболевание представленные в датасете чаще встречаются у тех людей, которые никогда не курили либо бросили. Это достаточно странно, возможно на это повлиял размер выборки.